Limiting the epidemic of HIV infection, whether by behavioral change or vaccine requires knowledge of how far and how fast it has spread. Prevention strategies must be targeted where they have the most effect, and vaccine trials can only be attempted in populations whose risk of HIV infection and progression has been characterized. Because the transmission of the virus is generally not directly observed, one must rely on indirect measures. This means that statistical models play a central role in our understanding of the epidemic. The excellent AIDS surveillance system established in 1982 by the Centers for Disease Control and Prevention, in conjunction with Health Departments around the country, is a valuable source of information. We propose to continue to develop analytical techniques needed to extract information from the surveillance system. Extensive work has been done in modelling the reporting lag and adjusting incidence to project accurately epidemic trends. Extensive modification of these methods is required to accommodate the new definition of AIDS, which is based on measured CD4 counts rather than clinical illness. Further, methods need to be developed to better model the component parts of the epidemic. The recent changes in the definition of AIDS have introduced yet another level of uncertainty in modeling the spread of the epidemic. Making inference about epidemic trends from surveillance data requires models of the natural history of marker processes and well as of HIV testing behavior. Simple data collection, using the surveillance model, can also provide information about the public health impact of treatment strategies. This will take on increasing importance as the number of therapies increases, and the need for large-scale postmarketing research becomes apparent. We propose to develop methods required for designing and analyzing studies (randomized and observational) that make use of simple data collection to answer important treatment questions. We also propose to continue to study techniques for making HIV screening tests more accurate and cheaper. Thus far we have proposed and investigated various novel ways of pooling and retesting blood samples. These have achieved, on paper, the twin aims of being more accurate and cheaper. In limited, preliminary field testing they have performed as predicted. Further research needs to be done in order to streamline them so that they may be acceptable in a production setting.